Why Centralized AI Architectures Will Never Work in Healthcare and Life Sciences
A few years ago, Jeff Dean at Google shared a figure illustrating why companies like Waymo continue to acquire more data to build more accurate autonomous driving algorithms and why consumer-facing companies like OpenAI continue to look for larger datasets.
State of the Art
in Medical Imaging AI
So what is the state of the art of AI in medical imaging? Recently, Parnav Rajpurhar and Matthew Lungren authored a survey paper, “The current and future state of AI interpretation of medical images.” This is an excerpt:
In considering the widespread adoption of AI algorithms in radiology, a critical question arises — will they work for all patients?
The models underlying specific AI applications are often not tested outside the setting in which they were trained, and even AI systems that receive FDA approval are rarely tested prospectively or in multiple clinical settings¹.
Very few randomized controlled trials have shown the safety and effectiveness of existing AI algorithms in radiology and the lack of real-world evaluation of AI systems can pose a substantial risk to patients and clinicians².
Challenges in Medical Imaging AI
In a nutshell, most AI models have been trained and tested on a relatively small sample of images (between 1,000 and 10,000 images), which is insufficient for real-world practice. After all, a car trained to drive in Palo Alto will likely not work in London.
The Need for Large-Scale,
Real-Time Data
There is plenty of data in clinics, research labs, and hospitals around the world – real-time data in healthcare machines (CT, ultrasound, MRI, blood analyzers, etc.) and offline data in PACS and EMRs.
The Case for a Distributed AI Infrastructure
Centralized architectures, which have powered ChatGPT, will not work in medicine. The data sizes are much larger, the demands for privacy are much higher, and the need for real-time results is much greater. Instead, we need a privacy-preserving, real-time distributed AI infrastructure.
So rather than move the data to the application, move the application to the data.
Timothy Chou
Lecturer @Stanford, Board Member @Teradata @Ooomnitza, Chairman @AlchemistAcc
3 min read · Jul 24, 2024
Centralized AI architectures will not work in Healthcare & Life Sciences
Centralized architectures, unfortunately, will not work for AI applications in medicine. The data sizes are much lar…
